diff --git a/CHANGES.txt b/CHANGES.txt
index e5bc1443f37..d594af0234a 100644
--- a/CHANGES.txt
+++ b/CHANGES.txt
@@ -34,6 +34,10 @@ Bug fixes
     implementations to be specified via the System property
     org.apache.lucene.store.FSDirectoryLockFactoryClass.  (Mike McCandless)
 
+ 4. LUCENE-821: The new single-norm-file introduded by LUCENE-756
+    failed to reduce the number of open descriptors since it was still
+    opened once per field with norms. (yonik)
+
 New features
 
  1. LUCENE-759: Added two n-gram-producing TokenFilters.
diff --git a/src/java/org/apache/lucene/index/SegmentReader.java b/src/java/org/apache/lucene/index/SegmentReader.java
index 6fa50f38401..34715a01a17 100644
--- a/src/java/org/apache/lucene/index/SegmentReader.java
+++ b/src/java/org/apache/lucene/index/SegmentReader.java
@@ -54,6 +54,9 @@ class SegmentReader extends IndexReader {
   IndexInput freqStream;
   IndexInput proxStream;
 
+  // optionally used for the .nrm file shared by multiple norms
+  private IndexInput singleNormStream;
+
   // Compound File Reader when based on a compound file segment
   CompoundFileReader cfsReader = null;
 
@@ -92,6 +95,17 @@ class SegmentReader extends IndexReader {
       }
       this.dirty = false;
     }
+
+    /** Closes the underlying IndexInput for this norm.
+     * It is still valid to access all other norm properties after close is called.
+     * @throws IOException
+     */
+    public void close() throws IOException {
+      if (in != null && in != singleNormStream) {
+        in.close();
+      }
+      in = null;
+    }
   }
 
   private Hashtable norms = new Hashtable();
@@ -457,6 +471,9 @@ class SegmentReader extends IndexReader {
       byte[] bytes = new byte[maxDoc()];
       norms(field, bytes, 0);
       norm.bytes = bytes;                         // cache it
+      // it's OK to close the underlying IndexInput as we have cached the
+      // norms and will never read them again.
+      norm.close();
     }
     return norm.bytes;
   }
@@ -473,6 +490,7 @@ class SegmentReader extends IndexReader {
     Norm norm = (Norm) norms.get(field);
     if (norm == null)                             // not an indexed field
       return;
+
     norm.dirty = true;                            // mark it dirty
     normsDirty = true;
 
@@ -494,13 +512,10 @@ class SegmentReader extends IndexReader {
       return;
     }
 
-    IndexInput normStream = (IndexInput) norm.in.clone();
-    try {                                         // read from disk
-      normStream.seek(norm.normSeek);
-      normStream.readBytes(bytes, offset, maxDoc());
-    } finally {
-      normStream.close();
-    }
+    // Read from disk.  norm.in may be shared across  multiple norms and
+    // should only be used in a synchronized context.
+    norm.in.seek(norm.normSeek);
+    norm.in.readBytes(bytes, offset, maxDoc());
   }
 
 
@@ -515,8 +530,27 @@ class SegmentReader extends IndexReader {
         if (!si.hasSeparateNorms(fi.number)) {
           d = cfsDir;
         }
-        long normSeek = (fileName.endsWith("." + IndexFileNames.NORMS_EXTENSION) ? nextNormSeek : 0);
-        norms.put(fi.name, new Norm(d.openInput(fileName), fi.number, normSeek));
+        
+        // singleNormFile means multiple norms share this file
+        boolean singleNormFile = fileName.endsWith("." + IndexFileNames.NORMS_EXTENSION);
+        IndexInput normInput = null;
+        long normSeek;
+
+        if (singleNormFile) {
+          normSeek = nextNormSeek;
+          if (singleNormStream==null) {
+            singleNormStream = d.openInput(fileName);
+          }
+          // All norms in the .nrm file can share a single IndexInput since
+          // they are only used in a synchronized context.
+          // If this were to change in the future, a clone could be done here.
+          normInput = singleNormStream;
+        } else {
+          normSeek = 0;
+          normInput = d.openInput(fileName);
+        }
+
+        norms.put(fi.name, new Norm(normInput, fi.number, normSeek));
         nextNormSeek += maxDoc; // increment also if some norms are separate
       }
     }
@@ -527,7 +561,11 @@ class SegmentReader extends IndexReader {
       Enumeration enumerator = norms.elements();
       while (enumerator.hasMoreElements()) {
         Norm norm = (Norm) enumerator.nextElement();
-        norm.in.close();
+        norm.close();
+      }
+      if (singleNormStream != null) {
+        singleNormStream.close();
+        singleNormStream = null;
       }
     }
   }
diff --git a/src/test/org/apache/lucene/search/TestMultiSearcher.java b/src/test/org/apache/lucene/search/TestMultiSearcher.java
index 948c88c9843..99e9ef7408c 100644
--- a/src/test/org/apache/lucene/search/TestMultiSearcher.java
+++ b/src/test/org/apache/lucene/search/TestMultiSearcher.java
@@ -149,7 +149,6 @@ public class TestMultiSearcher extends TestCase
             // no exception should happen at this point
             Document d = hits2.doc(i);
         }
-        mSearcher2.close();
 
         // test the subSearcher() method:
         Query subSearcherQuery = parser.parse("id:doc1");
@@ -161,6 +160,7 @@ public class TestMultiSearcher extends TestCase
         hits2 = mSearcher2.search(subSearcherQuery);
         assertEquals(1, hits2.length());
         assertEquals(1, mSearcher2.subSearcher(hits2.id(0)));   // hit from searchers2[1]
+        mSearcher2.close();
 
         //--------------------------------------------------------------------
         // scenario 3