Profiling showed that the reduced performance is due to semanticdb-find-table-for-include* functions getting pounded even though taking an average of 30ms, so I took a closer look at their implementation.  When I saw that evaluations are skipped over based on semanticdb-find-default-throttle values, I decided to try removing system from the throttle.  I didn't try this sooner because I thought a nil value for :system-include-path would be sufficient.  However, this improved performance to about a second or two.

Giving up Boost headers in exchange for a speedup from 45s to 2s is certainly worth it, so this solution will work for the time being.  Thanks for your help!