Neither of them have the IOPS needed for any kind modern applications.
While I agree with just about everything else you wrote; I do take some issue with this particular statement.
Modern applications meaning what exactly? If you're talking about big data then you're right... but you don't use a traditional SAN array for big data if you're smart; you use a proper object-based storage platform and scale it that way. Traditional SAN arrays suck for those workloads. But if you're talking VDI, SQL, web farms and general purpose virtualization then I'm afraid you're wrong.
Realistically when fully kitted out in SYNTHETIC BENCHMARKS, the "low-end" SC4020 will top out at over 400,000 IOPS. Yes, real-world will be a lot lower but you can comfortably expect 150K-200K IOPS with sub millisecond latency pushing about 6GB/s in real world operations. The SC8000 is more like 300K IOPS or thereabouts in those same synthetic benchmarks, but can push a lot more data (around 20GB/s) and still get in the 150K-200K IOPS range due to the fact that it can have more than one SAS chain while the SC4020 only has one. No, it's not the "million IOPS" you get advertised by EMC and the like... but in real world I have rarely seen even the might XTremeIO get anywhere close to that... 300-400K IOPS in real world is more realistic. Yes, faster... but not amazingly faster.
Besides, "modern applications" don't actually require near as much storage IOPS as most people think. I had a customer recently asking me about SANs that ran in excess of 150K IOPS for their new primary storage. We talked about a lot of expensive options (Violin, XTremeIO and so forth) but when we actually did some monitoring of their actual application load they were hovering around 60K IOPS. Even their peak load was under 75K IOPS so they ended up in a lower cost and more expandable solutions... and ironically given this topic here they ended up with a pair of SC8000's with Dell's "Live Volume" capabilities so they could move workloads around on the fly. So much cheaper than the alternatives and fit their actual workloads perfectly... not what some marketing guy was telling them they needed. Each SC8K is storing about a half petabyte each and can scale to about 2PB each.
I find this is incredibly common... most people don't actually know what their workloads are using so they tend to buy into the marketing hype being spouted by the high-end storage vendors. Then they end up overspending for their solution... yeah they have plenty of headroom and there's no doubt in my mind that they're getting the best performance they could possibly get out of it (because their applications don't really scale), but I've been amused when I've seen 5 year old arrays being decommed that aren't even close to their actual maximum performance because some salesdroid said they needed a million IOPS.
And to Dell's credit it's not like they're sitting on their laurels waiting. It's well known they have new stuff in the pipe for Compellent... probably for announcement at the next Dell World conference. I'll be interested to see what they do with their next generation hardware... the SC8K has been a fantastic platform but it is getting a bit long in the tooth. Having said that I think they could probably continue to do more with the SC8K because I've rarely seen one exceed ~10% of the CPU utilization.