It is terrible if there is no guaranteed combination of FBO attachments.
Yes, I find this terrible as well. It is however the unfortunate result of hardware and driver makers to resist the considerable efforts of commoditization by software providers and standard bodies. Being free of hardware, driver and operating system specific idiosyncrasies will remain an elusive ideal long as hardware companies cannot come together and commit to a consumer-friendly iron-cast baseline support of HW accelerated rendering.
It will make using FBO's almost useless for this audience.
If not, I think it will be too hard for authors to use FBO's, IMHO.
I feel condescended to right there. I (as a WebGL user) am fully capable of dealing with FBO validity as it is right now. No grand scheme of and matrix of support is required for me. All of dozens of WebGL framework makers can deal with FBOs. Anybody who's interested can deal with them right now if willing to write (or include) a dozen lines of code, and read a couple sentences about how their validity works. Checking for valid FBOs is not difficult.
I still really think we should have an extension explicitly enabling rendering to a floating point buffer. And I think we should provide the author with some guarantees about constructing an FBO. Together, I think we can make it easy for authors to use them and to have confidence that they'll work.
FBOs are working for me right now, and the guarantees that the validity makes about them is enough for me.