1. Real evidence of actual abuse by governments is pretty thin in most countries with regimes that are generally regarded as liberal democratic (small l and small d) unless having the capability to abuse is taken to be equivalent to abuse itself. Indeed, that probably is true even under most regimes commonly thought of as oppressive, although the range of behaviors such governments ignore may be quite limited.
2. Security of such data clearly is a risk, but one that admits mitigation. Various key escrow arrangements that have been suggested included provisions intended to reduce the risk and increase the difficulty and cost of escrow database compromise. Risk never is zero, and all one reasonably can require is that it be quantifiable and small enough.
3, 4. There is no real basis to argue that key escrow would make encryption more difficult or less convenient, as collection, indexing, and storage necessarily would be automated. It could present additional points of failure (or not) depending on whether failure to escrow would cause failure of the basic communication. Communication for commercial transactions may not be an important issue, as for legal trade it often will be possible to obtain the details from at least one of the participants by a suitable court order.
5. The obvious answer would be to provide the escrowed key, as national laws may require, to the government of the originator and recipients. In many or most cases, that would be at most one, since most communications do not cross national boundaries. That obviously would present issues, but for most people and organizations they would not necessarily be overly serious. Those who wish to shield activities from any of the governments demanding escrowed keys would have the most reason for concern, followed by those with reason for concern about security of one of the repositories against criminals or competitors. Increasing the number of repositories clearly would increase risk of control loss, however.
6. The customary government approach to refusal to participate would be criminalization, with a combination of detection procedures and penalties sufficient to discourage it.
The point of the original post was not to argue for key escrow, which has very little to recommend it, but to note that it would not be less private than plain text communication and might not add a great deal of risk, for most people, most of the time, compared to encrypted communication without escrow. Other approaches to law enforcement access include enforcing backdoored encryption systems, probably a much worse choice, and judicial warrants demanding delivery of the decrypted message by the originator or a recipient, depending on details of jurisdiction and treaty arrangements, with punishment for noncompliance.