RTI 4.5f vs 5.0 - Losing samples now

Gentlemen/Ladies

Here's the situation.   We have a home-brew DDS data recorder that when compiled using the 4.5f libraries will easily handle up to 200,000 108-byte data samples per second.   If we compile the exact same sotfware using the 5.0 libraries the data rate we can handle drops to 75,000 samples per second.  The QOS files are identical in both cases although it has been modified to accomodate the 5.0 version.

Other interesting points:

• The 4.5f version of our recorder does not use as much CPU as the 5.0 version.   80% vs. 100+% at 90,000 samples/second.
• The function/method 'loan_serialized_bufferI()' returns a data size of 108 bytes for the 4.5f version and 105 bytes for the 5.0 version.  The input data is the same in both cases.

We have tried some run-time code profiling using valgrind with the callgrind tool.  The results are inconclusive to the ignorant (us). 
Any help here would be greatly appreciated.