Paper accepted in journal, should I remove style file for arXiv preprint?

Legally, the style file at first only is the software which you use to generate your paper and thus there should not be any problems in most cases, for the same reasons that you do not need to ask Microsoft for permission each time you publish something generated with Word. I see three problem cases, however:

• If you use TeX, and ArXiv does not have the style file and thus you need to supply it, which would make it be published alongside your article. In this case you may be distributing a software that you are not allowed not distribute. In my experience, many style files are free to distribute though; have a look at it whether it contains some license.
• The style file includes a logo or similar (e.g., Royal Society Publishing’s style file does this), which in turn may not have a license that allows you to distribute it and also probably is a problem because of the next point.
• The style file makes your paper mention the journal’s name, publisher, or similar, which creates the false impression that your preprint is actually a publication of that journal.

Will the journal get mad? – Of course you need to regard the journal’s copyright agreement (which you accepted) with respect to publishing preprints in general or on ArXiv in specific (Sherpa Romeo is a database that helps you to find this out). But if the journal allows you to publish a preprint content-wise, I do not see any reasons why using its style file would enrage it, given that the above problem cases do not apply. Some journals even allow you publishing the journal’s version of the paper on ArXiv, e.g., the Physical Review series.

Finally, a stylistical argument: If the journal has a single-column layout but your paper has no long equations or something else that would make such a format a good idea, you may want to opt for a two-column layout for your preprint.

I think what matters in most cases is whether the journal staff (i.e., not just the external peer reviewers) made some corrections and improvements to the paper. If they did, you generally can't post the fruit of their labor on arXiv without violating the journal's rights. But you can post a version without those corrections and improvements -- and then you can even add additional other improvements of your own that were not in the journal's published version.