It's a cool hack, but really targets Facebook-level needs. For most users, the downside (eg compatibility issues) will far outweigh the benefits.
Also at least 90% of web apps are tiny or are startups that never made it, either way they don't need this for a long time, like you say.
My point is that for nearly all apps, executing PHP is very small part of the total request handling time, so even a significant improvement will have little overall effect. I bet many people excited by HipHop could get a much bigger performance gain by adding some simple caching)
From Wikipedia: "scalability is a desirable property of a system, a network, or a process, which indicates its ability to either handle growing amounts of work in a graceful manner or to be readily enlarged. For example, it can refer to the capability of a system to increase total throughput under an increased load when resources (typically hardware) are added."
Say you have 10 servers now and you expect business to grow by 10x in next year, so you put 6 months of dev. time into implementing caching. Then you get 10x the traffic in a month and you get V.C. investment as a result. You then have a problem, people are getting a bad experience and the development won't be done for another 5 months. Your CEO buys a rack full of servers, but you can't do anything with them because you don't have a scalable system. Alternatively you could find out after 6 months that you can't make that 10x performance jump after all.
Twitter had this problem a while back, essentially their DB server got overloaded but it was already a 8 CPU, 64GB machine and despite having lots of money they couldn't quickly solve the problem, because you couldn't really get a bigger system.
Facebook knows how to scale, they want to cut costs where they can to become (more) profitable.
scalability = (efficiency(N_2) - efficiency(N_1))/(N_2 - N_1)
Running it the developers workstation w/ APC and Memcache and local Percona MySQL, it maxed at 300 request/s. (It's just a workstation, remember). Anyway, there may be some observer effects here, but the profiler indicated that most the time was being spent on php functions and system calls.
So we might be in the 1% you're describing. Or _maybe_ it's a more than 1% after all.
The other application where this may be helpful is the cloud. In a system like this you have 100s or 1000s of sites on a single machine that may reach the combined traffic where this optimization can come into play.