In case you needed any more evidence that trying to use MathML in place of LaTeX only causes trouble, here's the BibTeX generated by Elsevier for one of their papers (curl -LH "Accept: application/x-bibtex"

And in case you're wondering what that big glob of junk in the title should have been: it's $\chi$. Or even χ, if you're feeling sloppy. If I'm criticizing Elsevier here, though, I have to also credit them with getting the diacritics in the names correct; many other sources wouldn't have. The doi/url duplication and the unnecessary publisher field are standard for dois obtained through this method and are not Elsevier's fault.

doi = {10.1016/j.ejc.2011.12.005},
url = {},
year = 2012,
month = {may},
publisher = {Elsevier {BV}},
volume = {33},
number = {4},
pages = {679--683},
author = {Zden{\v{e}}k Dvo{\v{r}}{\'{a}}k and Daniel Kr{\'{a}}l'},
title = {Classes of graphs with small rank decompositions are $\less$mml:math altimg="si1.gif" display="inline" overflow="scroll" xmlns:xocs="" xmlns:xs="{XMLSchema}" xmlns:xsi="{XMLSchema}-instance" xmlns="" xmlns:ja="" xmlns:mml="{MathML}" xmlns:tb="" xmlns:sb="" xmlns:ce="" xmlns:xlink="" xmlns:cals=""$\greater$$\less$mml:mi$\greater$$\upchi$$\less$/mml:mi$\greater$$\less$/mml:math$\greater$-bounded},
journal = {European Journal of Combinatorics}