Actually, I would describe this as a bug. Those lines should be filtered out. AsciiDoc Python happens to have the same bug, but I agree with you that it is not the behavior I expect.
Block-level comments fall somewhere between a preprocessor directive and a delimited block (very similar to document-level attribute entries). They are consumed by the line reader. This is important because we don't want to impact lines inside of a verbatim context such as a listing or literal block. However, within a table, I'd say that they should be filtered out.
Could you file an issue?