In the past this was not an issue, because stored numeric fields in Solr were only binary, but now they are consumed by the highlighter component.
If we dont fix highlighter I have no problem. We should add some checks in highlighter to prevent NumericFields from beeing highlighted (e.g. if the TokenStream does not support CTA, it should stop highlighting). The above hack was just faster to implement in Solr to not slow down merging (because index formats from 3.x and trunk are incompatible now, so trunk needs to be committed ASAP).
This is why I opened this issue. Just to record that its a little bit inconsistent. But of course your response is true, it may make more problems.