Like plesa said, it's stability and ratio matching what is important in application like this, not accuracy. On LT5400 it's awesome 7.5%, making it worse than usual thinfilm of 1c. What's is important there, that all four(or more) resistors in networks are embedded on single mica or ceramic substrate, with same kind of material and process, so even if they all drift, they will drift very closely together. So absolute ratio would stay particularly constant, or at least close to that.
That's why it does not really matter if it's 0.001% or 0.1% or 7.5% absolute accuracy tolerance there. Other thing is that more stable resistors are also usually more accurate, is a "side-effect" of how things manufactured. Sometimes you need absolute accuracy (for example for precision KVD or voltage divider, where value of resistors are important), sometimes you need just stability (then stuff like LT5400 will work good), sometimes you need both (then you need go custom networks, like Vishay foil networks, and prepare to pay gold for it).
Also worth a note, that it's not always tempco that important, but other things like power coefficient (resistor will drift with current flowing thru it, as resistive element will heatup itself), sensitivity to environment humidity and mechanicals, etc.