Some journals certainly do have HTML versions. For those that don't, the answer is probably a combination of institutional inertia and an eye on the bottom line.
In my "spare time" I am assistant editor on a very small, open access peer-reviewed journal. Even though it was born digital (meaning that we only publish online, and have for the entire decade or so of our existence), layout is done in Adobe InDesign1 and aims for a print-ready look. This is partly because the founders of the journal were used to that, and partly because so many of our readers actually print out and read articles, rather than reading them online (we know because we've surveyed).
Even though we are a small, new journal, I strongly suspect that many/most of the larger, more established journals also use traditional desktop publishing software rather than LaTeX, as they were (by definition) founded before LaTeX was a thing. I also suspect that most of these are specifically using InDesign, which currently dominates the professional market. Traditional desktop publishing may be part of their workflow even if authors submit papers in LaTeX.2
Over the years I have tried a few times to switch our journal over to a more onscreen-friendly format, but given the demands of our readers and board the only option is to add an online-version, not to do away with the print-friendly version. This is harder than it sounds. First, final-final edits aren't made until we are in InDesign, when they're easier to spot in proofs. That means that I don't have a raw file to work with, but rather need to output from the InDesign file if I want to be sure the text matches. This is not a very slick process (getting figures into the right place in the flow, for example, is very tricky). What that means is that I have to essentially type-set two different versions of each article.
Which brings us to the second major issue, the bottom line: I personally do not get paid extra for doing extra work to make a second version. You can probably guess how high my motivation is to do that gratis work when the first version is adequate for 99% of applications. Larger for-profit publishers would need to pay someone extra to do the extra work (people who do the grunt work of typesetting and the like are almost the only people who get paid for any part of the actual creation of academic articles); this again is likely to be a low priority if there is not an off-setting financial upside for them.
I do expect that there will be movement in this in the near future, however. For one thing, the technology for porting between one format and another is getting better/smarter, making the barrier to entry for HTML versions much lower. For another, there is a new focus at least in the US on accessibility of documents, and HTML is better for this because it is possible to embed more meta-data to facilitate technical accommodations. My brother works for a very large academic publisher, and he tells me that they also all work in InDesign. However, they are currently in the process of making HTML versions a higher priority, due to accessibility requirements.
1 When I started, the journal was actually still being formatted in Word, so InDesign is actually an improvement.
2 For example, even though Elsevier asks for submissions in LaTeX, they tell advertisers that they use a "PDF workflow" and a recent job posting for a Graphic Designer asks for "High level proficiency with Adobe Creative Suite (InDesign, Illustrator, Photoshop)" but doesn't mention LaTeX.