I'm still somewhat ambivalent about this – surely the EBS could simply ignore cases without an estimate as part of its analysis.
I've compromised in two ways:
- Enter an estimate of 1 minute.
- Create recurring 'Schedule Item' cases for things like reading email, but only for a limited period, like one month, and then enter an estimate for that item over that limited period.
The advantage of  is that it's obvious to other users that the estimate isn't intended to be accurate, but I'm sure this could wreak havoc with the EBS. I have some cases with hundreds of hours and I'd imagine that lopsided 'estimate histories' for those cases could heavily skew any analysis being performed by the EBS.
The advantage of  is that time spent reading emails, reviewing cases, etc. is being explicitly estimated over a long enough period that it's almost reasonable to make an estimate of those items. I've started using the 'elapsed time' from the case for the previous month when entering an estimate for the next recurring case for that item – i.e. if I spent 30 hours reading emails in February then I'll enter 35 hours for March (because of the difference in the number of business days between the two months).
The disadvantage of both compromises is that some (recurring) tasks simply can't be estimated, but we may want to still document the time we spend on them, and we may also not want to affect the FogBugz EBS.