Warts are things that would be done differently if we were to do it all over again.

The StorageInterface tiddler_written method is too useful to be left hanging as something to be overridden by monkey patching. It would be better to be able to register a tiddler written function. This has been fixed in 1.1.x.
HTMLPresenter is crufty and awkward. A header and footer function registration system would be useful.
Bag architecture and performance appears to be incorrect for how bags are actually being used. However the bag architecture is an explicit choice, an intentional constraint. The mismatch between expectations and design is a combination of bags being used in unexpected ways and poor documentation of bag concepts. There is probably middle ground that could be found where expectations and design and constraints are more aligned. See also Futures and 1.1.x.
if a tiddler title ends with ".ext" and ext is registered as a serialization extension, the individual tiddler can only be accessed in that particular representation (e.g. "index.html" as HTML)
The twanager commands for creating/editing bags, recipes and tiddlers (see e.g. twanager bag) take different kinds of input, sometimes matching serializations, sometimes not, resulting in confusion. Unfortunately the solution is not simple because the serializations are themselves confusing and the goal here is to make something useful for people to create entities from the command line. cdent thinks it would be ideal if someone or some ones were to create a or some twanager plugins which encapsulated the creation of the entities. This would help expose the right way to fix it all up.
Using the text serialization in a PUT of a tiddler can be confusing if the strict format is not followed: at least one header is required and if it is empty then you must have "<header>:<space>".