I rather objected to this bit:

Kolmogorov complexity has many powerful theoretical properties and is a central ingredient in the theory of universal artificial intelligence. Its most important property is that the complexity it assigns to strings and sequences does not depend too much on the choice of the universal Turing machine U.

...on page 32. Kolmogorov complexity and Solomonoff induction are highly language-specific. Almost all the discussions of Solomonoff induction I have seen gloss over this fact - with a comment along the lines of "oh, emulating another machine just takes up some fixed number of bits for the emulator" - so it doesn't make much difference.

IMHO, that is not acceptable. It does make a difference. IMO, if you choose a crappy, serial model of computation as a foundation for a metric, that choice needs more actively defending.