There's been considerable discussion of crossing improvement and grade separation costs in Times articles recently - for obvious reasons. Ventura County was really pretty blunt about it: they've wanted to grade separate the Rice crossing, and have tried to get money for it, but so far it's unobtanium. That crossing may be bad in terms of accident rate, but unless the County can find 25-50% of the cost (difficult for Ventura, which has no special transportation tax) a separation won't be happening soon.

A fairly simple overhead structure for a road crossing a railroad runs upward of $30 million. Where complications (such as a road along the tracks that needs a connection) occur it can hit $50m fast and break $100m in places with high land cost, or business or resident relocation needs. Even a good channelization and gates arrangement (suitable for quiet zone) for an arterial road can easily hit $10m. It ain't chicken feed, which is why with relatively tight money the crossings get scrutinized closely for cost-effectiveness.

The railroads normally don't have to make much if any contribution to road/rail grade separation costs. It's not their responsibility; the public agency that owns the road (alias state, county, local people, perhaps with a little Federal help from one of the most underfunded highway programs) has to cover the cost. If the railroad contributes, it's usually a token amount and/or some help with working out design and construction details. If it's a publicly owned railroad (like a commuter line), of course, some additional contribution will usually be extracted by the politicians. And it's generally understood that a high-speed line (essentially, anything routinely operating at more than 90 mph) should be fully separated when built or upgraded to that standard, but few if any private railroads are involved in those kinds of projects.