we have done some extensive testing on this. One of our clients had some severe issues with performance on Linux after a migration from solaris to linux (on the same hardware) the system load/utilization went up by a factor of 4-5 (server could easily handle 200 sessions with solaris and only 40-50 with linux). We never managed to find the root cause (or a solution) for this and migrated back to solaris.
But we tried to replicate this issue in the lab (without success). So what we did was this:
Install Solaris 10 and Linux on one hard drive each and also install SRS 5.3.1 (we also tried this with a few older versions). We set up kiosk sessions to connect to a windows terminal server (we did not bother to compare virtualbox perfoamce) and launch a web browser to a website with loads of flash and moving content. We had about 20 DTUs connect to this server and monitored the system load, utilization and so on of the Xnewt processes. At one point we also automatically started some audio stream to measure the performance of utaudiod. The result (for 2 different servers) was that linux and solaris are really close to each other in performance terms. Solaris showed a little less utilization, about 5-10% maybe. But really quite negligable.
Long story short: we have seen really poor performance with linux once but in all other cases linux performed almost as well as solaris. And we never compared the virtualbox hypervisors which might be yet another story.