(I don't know if there was a 7MHz clock that was divided by 2, or what, or if the thing could genuinely produce 2 pixels per cycle.)
4MHz was somewhat high compared to the 6502-based systems that made up some of its competition, but for many operations the Z80 doesn't run terribly efficiently. Opcode fetch takes 4 cycles and memory read/write takes 3 cycles, making the memory bandwidth of a 4MHz Z80 about that of a slighty-more-than-1MHz 6502. (The 6502 would read or write 1 byte every cycle.) This downplays the value of the Z80's register set and 16-bit operations, though; 6502 zero page is a bit of a poor substitute in many cases, because you lose a cycle from having to read the address byte.
(Compare INC $xx - 5 x 6502 cycles, 2 x opcode, 1 x read, 1 x dummy, 1 x write - with - with INC r - 4 x Z80 cycles, all opcode fetch and then it's done. On the other hand, ADC A,n takes the Z80 8 cycles, because it has to read 2 opcode bytes - something the 6502 will get through in 2.)
I've spent a lot more time using the 6502, but I like both. They are represent two very different styles. The Z80 is a lot more practical in many respects, with its 16-bit registers and relocatable stack pointer. Some 6502-style register-indexed indirect addressing modes wouldn't have gone amiss, though...
For instance, the Z80's fastest instructions took four cycles, where the 6502's fastest instructions took two cycles.
The Z80's slowest instructions took 23 cycles, and the 6502's slowest took seven cycles.
And even that comparison is over simplified.
Also it has a real actual genuine soul connected to a higher plane of existence that you communed with every time you used it.