This is actually a complicated electrical question dealing with capacitance, crosstalk, signal loss, and a number of other factors. That stuff makes my brain hurt though, so as a rule of thumb let's just say that passive cables should be limited to the shortest length, and should always be kept under the specification limits.
As a practical matter a cable slightly longer than the specification limits will probably still work, but it would be Bad And Wrong of you to attempt such a thing, and the engineer who wrote the specification will find you and kill you in your sleep.
So for a configuration of:
[Controller]<--SAS Cable--><--SATA Cable-->[SATA Drive]
or
[Controller]<--SAS-TO-SATA Cable-->[SATA Drive]
the two total cable length between the controller and drive should be limited to a maximum length of 1 meter (the SATA limit).
One meter is a REALLY FREAKN' LONG CABLE -- I would be hard pressed to find a cable routing that would exceed that in most servers/racks. If you're in that situation with external cabling in your rack the usual suggestion is to move a machine to reduce the distance.
When active components are involved (as they would be in your case) things are a little different.
The length of any passive cable is still limited, but active components (at least partially) reset the distance counter, so in a configuration like this:
[Controller]<--ESATA Cable-->[JBOD SAS Backplane]<==SAS Cables==>[SAS Disks]
the ESATA cable length is limited to 2 meters (the ESATA specification limit), and the cables betwee the JBOD's backplane and the SAS drives is limited to whatever the SAS limit is (I think it's 3 meters but don't quote me on that).
As a practical matter you can't just keep cabling things together for an infinitely long chain (even if you're sticking an active component in at each distance limit) - eventually the delays introduced by the cabling and the hardware would start to cause problems. Whether you hit other practical limits before that (like controller bandwidth saturation).
