Not sure it's any better than a used big brand device from ebay, as for those I can usually easily buy warranty or support for not much money.
Perhaps I should have clarified that in my last paragraph; repair plans are easily amortized if you use the equipment professionally, it is a lot of money for hobbyist or personal purchases. [/quote]
It's not expensive, in fact, these repair agreements are often surprisingly cheap.
In your example, the repair plan will set you back $1.1k on your 33522B (1/3 of the price, according to Keysight).
Not sure what makes you think a repair agreement for a $3600 device is $1.1k (which would be ludicrous):
https://service.keysight.com/infoline/public/product-service.aspx?laf=mya&pn=33522B&lc=eng&cc=USA single year costs a staggering $72. I'm in the UK, and I just paid a bit under £300 for 3yrs repair agreements for my 33522B and my E5810A GPIB LAN controller, both bought very cheaply from the US via ebay.
One could argue that, if you deem that brand so dependable, why spend so much in a repair plan? O0
Because, as an engineer, I am well aware that any piece of complex electronics can fail, and the repair agreement is a very cheap insurance against that risk. That means I got the same protection as with a brand new device, while paying an overall much lower price.
The second hand big brand instrument will also very likely perform much better (...)
That is where you lose me. While I agree with your overall comments about the repairability, brand alone does not mean dependability and performance anymore. The marketplace is changing, and it is not unusual to see comparably priced equipment losing ground in features, "featured bugs" and even build quality.
I deal with a lot of test equipment as part of my job, and aside from the LeCroy WaveRunner Xi (which has a pretty poor build quality) I can't recall when I came across a big brand instrument with inferior build quality.
As to features, B-brands like Rigol tend to add features in a check box ticking exercise without thinking about the implementation. The best example is the sample memory of their scopes. 140M in the DS6000 sounds like a lot, it's much more than the big brands offer, and it looks really great on the spec sheet. But the sad reality is that you can't do a lot with it as there are no search or analysis tools to find stuff like runts or glitches in memory. And manually searching through 140M of recordings is absolutely painful. Plus, despite the large memory, FFT is limited to
a measly 2048 points according to Rigol (although the owner of the DS000 told me it has been increased to 16k points in a recent firmware update), which is really nothing else than embarrassing (most DSOs in the late '90s had better FFT). Big brand scopes in the same class come with less sample memory but they do have tools to make better use of it, plus FFT uses a lot more points, and in the lab that's a much bigger advantage than having a huge sample memory with no tools. Other features (math, trigger, measurements) of the DS6000 are pretty basic, more akin to what you'll find in modern bottom-of-the-barrel scopes.
The other problem with B-brands is that they often tend to implement features in a half-working state. Just check the threads for any Rigol gear (DS2000, DS1000z, DS4000), almost all of their products come with non-working or half-working features, plus a ton of other bugs.
Big brands scopes may not come with some sticker features as a very large sample memory, but they generally come with better implementations of things that don't look sexy on the spec sheet but make a scope much more useful on the bench. They will generally bring a product on the market when its in a mature state, and any bugs that slipped through testing are usually fixed pretty quickly.