The challenge here is that the slide backends are written "from scratch". That means they have to reimplement all the conversion logic in core faithfully or else you'll observe different behavior. Even further complicating the issue is the fact that AsciidocFX has to use yet another set of converters since the Slim templates can't be loaded in Asciidoctor.js.

Over time, all of these little differences will get smoothed out, but for now they should be treated as bugs / improvements and filed in the issue tracker. For the reveal.js backend, you'd file those issues here: https://github.com/asciidoctor/asciidoctor-reveal.js

@ted, you don't need the AsciiDoc style on a cell to use inline images. The AsciiDoc style allows you to use block syntax (blocks and block images) in a table cell. Anything that works in a paragraph will work in a plain AsciiDoc table cell (I just updated the User Manual to make this more clear).

Speaking of the $subject, I've been wanting to explore the idea of using a CSS grid instead of a table to layout content inside tables marked with the "grid-layout" role. There's no reason that an AsciiDoc table has to map to an HTML table. (or we can use nested blocks of some sort). That's the beauty of having a converter framework in the processor. For now, you'd need to create your own templates to do this, but I'd eventually like to see this idea supported in core. (Btw, Sean has been implementing a related idea for achieving a tab-based layout).

Cheers,

-Dan