I dont recall exactly which TCPs this broke, but I do remember the implementations for VXWorks, some thirdparty TCP for mainframe OSs (IBM 370), and Solaris were often the culprit, and less often there was a problem with a Windows TCP (remember there are different releases to test of each. If you don't, your customers will). While I realize you may not care about those servers, you really don't want to make changes to the client that are incompatible.
I'd suggest checking with the SPDY guys. They've done tons of testing of various latency improvements.
> 9.1.2 Idempotent Methods
> The methods GET, HEAD,
> PUT and DELETE share this property. Also, the methods OPTIONS and
> TRACE SHOULD NOT have side effects, and so are inherently idempotent.
RFC 2119
> 4. SHOULD NOT
> This phrase, or the phrase "NOT RECOMMENDED" mean that
> there may exist valid reasons in particular circumstances when the
> particular behavior is acceptable or even useful, but the full
> implications should be understood and the case carefully weighed
> before implementing any behavior described with this label.
What could possibly go wrong?
Obviously we aren't going to be able to blindly enable this optimisation without making sure our apps meet the assumptions first.