That data looks more like noise to my trained eye - there's no relation between adjacent points, and how likely is it that a card at full pelt will happen to draw more-then-less-then-more-then-less-then-more each millisceond and so on to the tune of +/- 100W or more? Looks to me like someone plugged their probes into their oscilloscope and took the numbers coming out as gospel.
That aside, cards might spike but in the same way for good PSUs the spec shouldn't be taken as a hard limit:
http://hardwareinsights.com/wp/seasonic-g-450-review/2/ - Seasonic 450W found capable of supplying a steady 540W
http://hardwareinsights.com/wp/thermaltake-tpg-650w-review/4/ - Thermaltake 650W found capable of 800W
http://hardwareinsights.com/wp/corsair-vs450-review/4/ - Corsair 450W capable of 500W
http://hardwareinsights.com/wp/power-logic-magnum-pro-315-power-supply-review/5/ - Cheap 315W does 320W and goes bang when overrun.
Anandtech gave a total average system consumption at high demand of < 400W (http://anandtech.com/show/7481/the-amd-radeon-r9-290-review/15) for an R9 290 system. Even believing the 1 ms spike of an extra 100W would mean < 500W draw. The rest of the system would have also have to spike from 150W to 200W during that
exact same millisecond (1ms either side and the GPU draw is likely down below 200W according to those plots) for the total to hit 550W, and even then a quality PSU will have an extra 50-100W overhead to cope.
The 550W powering my R9 290 hasn't gone pop yet, despite all the apparent spikes in power draw it must've seen. If it does though, I'll be sure to report it in the interests of evenhandedness.