Skipping over the constraints on a single score (for the moment), it seems like the hotspot in your question is getting the 90th percentile of scores across a range of leads. Is that indeed what you are trying to extract?
As Sanford suggests, this might be a fairly complex question to run in Marketo.
Sales Insight lets you adjust this threshold in Admin > MSI. But that's more about flames/stars in SFDC, than a real threshold. And since you want this to be a rolling 3 months, how do you determine this?
- Get a predictive vendor tool
- do an extract and analysis on behaviors and score those within a 90 day period, using an Inactivity batch to lower the score after 90 days of inactivity (not really aligned to your desires those, but closer).
- but that data analysis would get you an idea of the 10% of most active and model them.
- you'd need a regular review.
I wouldn't look at Channel as much as which behaviors lead to MQL or SQL.