Why don't miners get boiled to death at $4$ km deep?
As noted in CountTo10's answer, the main answer is simple - miners don't "boil" because the mines use suitable cooling and ventilation equipment, plain and simple.
That said, there is a contradiction, at least if you go only by Wikipedia and explicitly ignore its caveats. The Wikipedia page for the Mponeng gold mine makes the maximum rock temperature at 66 °C, and if all you read from the Wikipedia page on the geothermal gradient is the stuff in bold, then yes, a 25 °C/km gradient over a 4 km depth would give you 100 °C on top of the surface temperature.
However, the actual text in that page reads
Away from tectonic plate boundaries, it is about 25 °C per km of depth (1 °F per 70 feet of depth) near the surface in most of the world.
and makes it clear that there can be local variations. With that in mind even some very mild digging turns up this map of the geothermal heat flow in South Africa:
(Taken from S. Afr. J. Sci. 110 no. 3-4, p. 1 (2014).) This makes it clear that the Mponeng mine is right on top of a cold spot in the Wits basin.
The stated heat flows are not enough to reconstruct the thermal gradient (you need the thermal conductivity for that), and I'm not going to go on an expedition for fully trustworthy sources for that gradient. However, some more cursory digging unearthed this source, which looks reasonable (if not particularly scientific), and which claims that
mining at these depths is only feasible in South Africa’s Wits Basin due to a relatively low geothermal gradient (nine degrees Celcius/km) and the presence of gold reefs in hard competent country rocks.
This is enough of an agreement to call it a day. Backtracking a 9 °C/km gradient over 4 km gives a ~36 °C difference, and taking that away from the 66 °C (maximal!) rock temperature in the mine gives a ~30 °C average surface temperature. This is relatively high, but it is within a reasonable envelope, and there's plenty of leeway on the numbers (e.g. making the gradient 10 or 11 °C/km) to take away any glaring contradictions.
There is this very short account on Wikipedia, which seems to cover it.
Mponeng Gold Mine
The temperature of the rock reaches 66 °C (151 °F), and the mine pumps slurry ice underground to cool the tunnel air below 30 °C (86 °F). A mixture of concrete, water, and rock is packed into excavated areas, which further acts as an insulator. Tunnel walls are secured by flexible shotcrete reinforced with steel fibers, which is further held in place by diamond-mesh netting.
From Wall Street Journal
It takes 6,000 tons of ice a day to keep Mponeng's deepest levels at a bearable 83 [°F] degrees. They make the ice in a surface plant, then mix it with salt to create a slush that can be pumped down to underground reservoirs. There, giant fans pass air over the coolant and push the chilled air further down, into the mining tunnels. Cool air goes down at a temperature of 37 [°F] and comes back, heated up by the rock, at 86 [°F]. I walked past one of these hot air returns—a black, growling tunnel that exhaled rank air from the bottom levels. (Emphasis Mine)
The gradient you're quoting is an approximate figure that applies to most places. There could be local variations, both in terms of where on the globe the mine is and in terms of the gradient not being exactly constant with depth. It's not so surprising that this one mine is off prediction at a depth of 1 km, which is only about 1% of the lithosphere. One might speculate that having an air pocket (the mine) would passively affect the temperature of the local rock as well before the active cooling system is considered.
From wikipedia:
Geothermal gradient is the rate of increasing temperature with respect to increasing depth in the Earth's interior. Away from tectonic plate boundaries, it is about 25 °C per km of depth (1 °F per 70 feet of depth) near the surface in most of the world.