First of all: I'm not very familiar with screen readers and associated markup. I just use VoiceOver to see how accessible a page is, but that's it.

One thing I noticed is that VoiceOver reads the structure of a table when going through the cells. This can be observed e.g. with the vector skin, or more generally for tables on other sites. For instance, if I go to this section with Vector and let VoiceOver read the first table, it will read:

election, row heading, column heading, row 1 column 1; table start, 8 rows 7 columns
house, column heading, row 1 column 2
constituency, column heading, row 1 column 3

etc. Note: these may not be 100% accurate. Because the language of my device is set to Italian, VoiceOver uses a weird mix of Italian and English when reading the interface; I've translated some words.

But if I open the same page with Minerva and let VoiceOver read the same table again, it'll say:

election house constituency party votes result

That is, it doesn't say anything about the table structure, and it reads all the headers at once without pauses.

Another thing I noticed is that this only seems to be the case for simple tables; for instance, VoiceOver reads the structure of the table in https://en.wikipedia.org/wiki/Quenya#Numerals with Vector as well as Minerva.