First, you need to define "Bad Leads" which could be people w/o data, people who haven't been updated in 18 months, etc...people with no activity since 6 months ago...
Then you can run a batch campaign to delete them from SFDC and/or Marketo. You should probably archive them by exporting that list FIRST...just in case. Be sure this isn't going to throw off your reports.
For Unsubscribes you can send them to a special Queue. But you might want to be careful. Just because they unsub doesn't mean they are a bad prospect for sales. Usually people deduct points for unsubs.
You should also monitor repeated hard bounces and place those on to Email Invalid or Bad Leads eventually because they will start to affect your email rep.
Following Josh's post: there is not an archive function per se in Marketo.
In general, I've found the two reasons to delete leads are to save on the cost of Marketo and remove absolutely useless dirty data.
If the email address is properly formed, you should archive them by using some lead status such as "Archived" "Bad Data" "Bad Lead" etc. In Salesforce, this can be exposed to the Sales team so they can update the status; does SAP have a similar setup?
We have several ways to flag these and isolate them. If our sales reps determine that the lead is "Mickey Mouse", they change the Lead Status to "Bogus Info" and if the lead is no longer at that company, they change the Lead Status to "Gone". In both cases, this triggers a smart campagin to change their lead revenue stage to "Bogus Info". We also periodically move leads with Email Invalid and that have had no activity in the last 6 months to the "Bogus Info" revenue stage. We haven't done it yet, but they be exported (all columns) and deleted as I've never seen a lead in the Bogus Info revenue stage come back to life.
Im trying to do a data cleanse. I want to speed up the process of our smartlists/programs and have a more accurate representation of true marketable leads. It's not to lower the cost of marketo, just to have a more accurate uninflated database. Thanks for the suggestions