I suppose it makes sense with moving time window Splunk has to keep track of the window and events fitting that window. If you don't include the current event Splunk doesn't know how many previous events it has to keep and include in your calculations.
If you have a fixed window expressed in number of events - that's easy - Splunk has always remember last n events to calculate your aggregation. But in case of time window it would make Splunk have to remember much more events that are used to calculate the stats in case they don't "fall out" of the window in case of next event. So it's simply easiest to forbid using use_current=f probably.
I suppose you could do some ugly hacks like streamstatsing lists of values and manually calculating your aggregations but that would probably be horribly inefficient.
I must say that I don't see a use case. What that would even mean "5-minute window without current event" - A window of 5 minutes looking back from the previous remembered event? Or a window of 5 minutes looking back from current event but without using current event's value?
In the latter case you could simply do a "half automatic" calculations - for example with an average, you could just streamstats sum and count, then substract the current event's value from the sum and use count (or count-1) to calculate average. For more sophisticated aggregations of course you'd need to be a bit more creative.
But the former case - it doesn't differ from use_current=t if you're just aggregating from previous event backwards.
Maybe there's something more to this case you're not telling us and it can be done in yet another way.
