A collection of small optimizations have been shown to improve the performance of simple counts about 3x in testing. They include:

• caching a BtreeBucket retrieved from a DiskLoc in IntervalBtreeCursor
• a simplified implementation for checking if the read lock should be yielded in count.cpp
• reducing the number of calls to Cursor interface functions in count.cpp and IntervalBtreeCursor
• a new Cursor::acceptCurrent() function that avoids a call to currLoc() and avoids virtual calls in some cases