Cygwin currently does not support named pipes/FIFOs. refdbd uses these to implement a child->parent messaging after the application server has forked. On Cygwin, this messaging is emulated with a temporary file. This is a bit slower (and a lot kludgier and error-prone) than a named pipe and almost certainly is a security hole.
As the Cygwin library has to translate all calls to Unix system functions to native Windows function calls, there is a small performance penalty in comparison to native Windows or Linux/Unix applications. My general impression however is that this does not affect the overall performance as most time seems to be spent during the database search itself (i.e. inside MySQL). You may gain some milliseconds by running the application server on a Linux/Unix box, though.
Building refdb on Solaris works only with gcc, not with Solaris' own cc.
Of course a major problem is the lack of support for more platforms. RefDB may work on a variety of platforms besides those mentioned above, but unless you tell me so, I'll never know!