Best approach to maximize performance on semi additive measure

Hi

I would like to know if someone has some experience to share...

I have a cube with semi additive measure (bank account balance). The measure is defined by calculated member like:

<Measure name="SumBalance" column="daily_balance" visible="false" aggregator="sum" formatString="#,##0; -#,##0"/>
<CalculatedMember name="Balance" dimension="Measures">
<Formula>CASE [Period].Level WHEN [Period].[Days] THEN [Measures].[SumBalance].Value ELSE Avg(Descendants([Period].CurrentMember, [Period].[Days]), [Measures].[SumBalance].Value) END</Formula>
</CalculatedMember>

It is working fine but since the data is filled every day for multiple companies. it starts to become a little slow. For example, I made a CDE dashboard with several charts and it takes about 20 seconds to render, for each company selected. After that, data is in cache and it gets fast enough. I tried many combinations with aggregate tables but this just doesn't help too much (as aggregate tables contain only base measures) so it seems like dynamically populating cache is the best approach. The little problem here is that cube is filled with data by our custom online form, cache is programatically cleared after each fill (so that JPivot/Saiku views hold accurate data) so I am not sure if populating the cache after each fill is a clever idea.

Anyway, if someone is dealing with something similar, I would appreciate any advice

Thx, dejan