It has also been proposed that a Java implementation is all that is needed, and building a number of platform-dependant web texture engines is a backwards approach.
While in the long term, this approach may be a feasible one, at the moment it is not practical due to performance issued with Java. No doubt in the long term we may find that existing web browsers are overtaken by browsers build entirely in Java. Sun's own HotJava is supposed to be quite respectable these days.
The reasons why an uptake of Java is not going to happen too rapidly are not entirely due to the established code base. Java will need to have been around for some time before its performance improves and it will not ever reach the point where it becomes of a similar order to that of native code, (even when running on JIT compilers or dedicated Sun chips). This is due to the cautious nature of the language and all the checking that is performed while running programs.
As algorithmic texture generation is computationally expensive, a Java implementation of it would currently incur too large a performance hit to be practical on anything but high end machines. As time passes, this position will improve and Java will become a more practical solution.
For Java information, the classical sources seem to have become: