I think to find out you'd need to at least measure on the same OS and hardware. A lot of things happen between the physical hardware and the kernel socket layer and those might be different between operating systems.

Some of the stuff is difficult and time-consuming because "messaging" is generic enough to be configured and used differently by different users.

Obviously you can cut away some of the choicesright of the bat if you are worried about support for some OS (Like you have to ship on HP UX), or you need to have durability and acknowledgement and high availability, or you want a project with a certain level of maturity and stability and so on. That cuts the number of systems to test.

Then of course there are things like, well how do they handle concurrency. Just because a single producer and single consumer can do 500K messages per second (which maybe a small benchmark on a co-workers laptop will show), doesn't mean that the whole thing won't blow up and crash in a burning mess if there are 1000 consumers and producers.