Another reason is that writing an efficient gc is a very complicated software engineering task. A concurrent continuously compacting pause-less gc is the cold fusion of the software world (i'm half-kidding because a ccc gc does work in practice). :) So Dijkstra could propose it, but he couldn't (afaik) code a high-performance proof of concept. If there was a toughness rating scale for engineering problems, then writing a CRUD app in Django would be a one and a good gc a 10.

The problem is that you have really complicated system (the gc) meeting low-level language. You need to write it in C or C++ for performance and for direct memory access. But it's a language that isn't suitable for complex algorithms so if you could choose, you'd rather pick Lisp or something. It wouldn't surprise me if it takes Google a very long time to get their awesome gc that they are planning for Go up and running even if they throw heaps of engineers at the problem.