Add background layer to processed images?

Some time ago we agreed to include all information that belongs to a specific observation with the science data in a multi-extension FITS file (MEF). Therefore most data products have these three layers:

• SCI, the science data.
• VAR, the variance. (The variance is easier to use than the standard deviation because the pipeline is continuously adding to the variance layer, while a standard deviation would require squaring and square-rooting.)
• DQ, data quality, with hot/cold/etc pixels.

We can also add a background layer (e.g. BG to this), so it is always clear what background is used. This currently described in the add-background-layer recipe step. The science data should be considered background subtracted, because this probably makes it easier to do later steps.

Questions:

• Are there reasons to not include the background layer? E.g. keep the background around in a separate file instead.
• Are there reasons to not subtract the background from the science layer? E.g. what does SExtractor expect?
• Should the background have it's own variance layer or would one variance layer be sufficient? (Leaning towards a single VAR layer.)