Starting with 2bec281, WebServer errors out with EAGAIN when sending responses larger than a few hundred bytes.

The possible mitigations to this are:

• Reverting the patch and instead lifting the restriction on zero-sized writes.
• Implementing a cache kernel-side, which makes Nagle's algorithm transparent to the user (and provides us with the ability to consolidate packets?).
• Catching EAGAIN in Core::Socket and retrying automatically (potentially obstructs other, non-Nagle EAGAIN errors)
• Catching EAGAIN in WebServer and retrying automatically (least impact, but would obviously need to be done for each affected application)

Collecting behavior samples from other implementations of Nagle's algorithm would presumably also help.