Btrfs: broken file system design (was Unbound(?) internal fragmentation in Btrfs)

In computer science I recall sort algorithms that are almost always fast, but in the worse cases horribly slow. One is recommended to randomise the sequence of one's input data before using one of these, thereby destroying pre-existing mal-order. However, there is no proof that one can't randomise one's data into a pessimal order - indeed, a trivial proof that one can. It just becomes thermodynamically unlikely as the number of items increases (and for the case of small N, then even pesimally slow is tolerably fast). In this case one can calculate the probabilities.

There are a large number of algorithms which do not have provable bounds, or which have extremely undesirable provable bounds, but which work fine in practice. If btrfs is one of these, do we need to worry if it has theoretical weaknesses? Perhaps, if it is possible to run an attack on a filesystem that fragments it into uselessness even after deletion of the files created by the attacker. Certainly, if it is possible that real-world usage may accidentally arrive in this unfortunate state. Not at all, if such an attack is only theoretically possible, but not realizable in the real world by a party lacking access to the filesystem's internal structures (or more accurately, realisable in the real world only with a thermodynamically small probability for a realistically large filesystem)

Which is it? It may be that even this question is not amenable to theory, and will have to be resolved by real-world testing.