While there are equivalents available for interpreted/CGI languages (e.g. opcode caches, memcached), there is still a significant performance advantage. With a warm cache, most pages can be written out in under 5ms, and the busiest in under 50.
For some perspective, the time needed to compose the busiest view (threaded or split-view) halved when I moved from naive string concatenation to an optimized appender.
Yes, there are way too many things hard-coded at the moment. I want to refactor all of it out into templates, config files and library code. There's apparently some interest of adopting some code into D's standard library, too.