diff --git a/lucene/CHANGES.txt b/lucene/CHANGES.txt
index c44440e7e2b..ecdf5632acf 100644
--- a/lucene/CHANGES.txt
+++ b/lucene/CHANGES.txt
@@ -385,6 +385,10 @@ Optimizations
 
 * LUCENE-4598: PayloadIterator no longer uses top-level IndexReader to iterate on the
   posting's payload. (Shai Erera, Michael McCandless)
+
+* LUCENE-4661: Drop default maxThreadCount to 1 and maxMergeCount to 2
+  in ConcurrentMergeScheduler, for faster merge performance on
+  spinning-magnet drives (Mike McCandless)
   
 Documentation
 
diff --git a/lucene/core/src/java/org/apache/lucene/index/ConcurrentMergeScheduler.java b/lucene/core/src/java/org/apache/lucene/index/ConcurrentMergeScheduler.java
index 197846d9d93..60b655c5daa 100644
--- a/lucene/core/src/java/org/apache/lucene/index/ConcurrentMergeScheduler.java
+++ b/lucene/core/src/java/org/apache/lucene/index/ConcurrentMergeScheduler.java
@@ -55,14 +55,15 @@ public class ConcurrentMergeScheduler extends MergeScheduler {
   // forcefully pause the larger ones, letting the smaller
   // ones run, up until maxMergeCount merges at which point
   // we forcefully pause incoming threads (that presumably
-  // are the ones causing so much merging).  We dynamically
-  // default this from 1 to 3, depending on how many cores
-  // you have:
-  private int maxThreadCount = Math.max(1, Math.min(3, Runtime.getRuntime().availableProcessors()/2));
+  // are the ones causing so much merging).  We default to 1
+  // here: tests on spinning-magnet drives showed slower
+  // indexing perf if more than one merge thread runs at
+  // once (though on an SSD it was faster):
+  private int maxThreadCount = 1;
 
   // Max number of merges we accept before forcefully
   // throttling the incoming threads
-  private int maxMergeCount = maxThreadCount+2;
+  private int maxMergeCount = 2;
 
   /** {@link Directory} that holds the index. */
   protected Directory dir;