Toyota Landcruiser Prado long crank times, intermittent.

I have a 1997 Prado here with the 3.0 1KZ-TE engine. A mechanic has done quite a bit to it because of long
crank times when it is cold. I have checked the [new, correct] glow plugs individually drawing 10 amps each and
the glows draw 40 amps all together so nothing wrong there.
The mechanic has had the fuel pump overhauled twice! The injectors have been overhauled. A non return valve fitted to the fuel
inlet. EGR blanked off. He tried running a direct fuel line bypassing everything else.
Now I have it. I have scoped the ignition and battery positive and ground at the ECU and the ECT sender, fuel timing control and fuel
quantity control [which go through the ECU relay proving it has good ignition].
It's almost like the engine timing is mechanically changing and I'm wondering if this is possible, that something driving the pump
is losing its position.
Yesterday at zero degrees Celsius [32F?] it would crank for 10 second or more and not start, billowing white smoke. I sprayed some "electrical cleaner" into the intake and it fired up and drove well for 10km then i returned to the workshop. Once it was hot it started easily and progressively took longer to start as it cooled down. I double/triple checked the glow plug current draw and it is good.
This morning at -1 degrees Celsius it started on the first try within 2 seconds so it can't be a glow issue. It gave a puff of black smoke after starting. It started well several times then got worse again. The strange part is that sometimes you can crank it for ages and it doesn't start then just flick the key off and on again and it starts easily - like a petrol car I once had with centrifugal weights in the distributor sticking occasionally and throwing the timing out.
Any suggestions would be appreciated

I'd been checking for a blown head gasket or water entering the intake somehow.

Thanks for the reply, the mechanic eventually took the vehicle away with the problem mostly unresolved. [no blown head gasket]
It was improving over time somehow.
The weird 'clue' was that several times it would start quickly first thing in the morning at -2 degrees Celsius [33F?] and after running for
a minute and turning it off, 15 minutes later it would take 8-10 seconds of cranking to start. Another 15 minutes later it would take 14-15 seconds to start [shutting it down right after it starts each time after the first start].