Beaglebone Black can do it, but 100MSps is a pretty slow scope.
https://github.com/abhishek-kakkar/BeagleLogic
That is a modern example of the DMA method using integrated memory. The ARM processor in the BeagleBone includes a pair of 200 MHz 32 bit PRUs (PRUSS = Programmable Realtime Unit SubSystem) which can be programmed to operate effectively as a DMA controller or many other things:
http://processors.wiki.ti.com/index.php/Programmable_Realtime_Unit_SubsystemThat kind of processor seems ideal for something like a FPGA-less DSO although it would be limited to parallel interface ADCs. That is not a problem though because existing parallel interface ADCs can operate at 100 MS/s and faster. Some of them have bandwidths of 1 GHz and above (LTC2242 and at 12 bits!) so you could make quite a useful DSO although it would have to support equivalent time sampling to take advantage of that kind of bandwidth.
Supporting more than one input channel might be a problem but the processors are cheap (and the ADCs are not) so use one per ADC.