The HTML table model allows a better preparation of table structure for (especially screen reader) accessibility when transforming to HTML (and, by analogy, probably to PDF and other digitally accessed output formats too, though I haven't studied this yet).
The CALS table model does not come anywhere near the abilities of the HTML model in associating data cells with table headers.
Especially in complex table setups, a combined use of the HTML-model id and headers attributes is excellent for demystifying the structure of table data - where CALS does not provide any infrastructure that could be used to generate such structuring via XSLT.
Thus, including the HTML table model would cater to accessibility requirements in all publishing contexts transforming docBook content to be accessed online as HTML - and probably also other formats.
Techniques for WCAG2.0: http://www.w3.org/TR/WCAG20-TECHS/H43.html
W3C research on complex tables: http://www.w3.org/html/wg/wiki/ComplexTables
WebAim Guide for accessible data tables: http://webaim.org/techniques/tables/data